bool -> np.bool_ casting for any bool

[Yaroslav Halchenko]

I wonder if that is intentional change introduced in 2.0.0 since in
1.3.0 it is all good: consider following snippet:

import h5py, numpy as np
hdf = h5py.File('/tmp/1.hdf5', 'w')
v, vnp = True, np.bool_(True)
print "original types ", type(v), type(vnp)
hdf.create_dataset("1", None, None, v)
hdf.create_dataset("2", None, None, vnp)
# get it back bow
v_ = hdf.get("1")[()]
vnp_ = hdf.get("2")[()]
print "read back type ", type(v_), type(vnp_)

on 1.3.0 it returns:

original types <type 'bool'> <type 'numpy.bool_'>
read back type <type 'bool'> <type 'bool'>

and on 2.0.0:

original types <type 'bool'> <type 'numpy.bool_'>
read back type <type 'numpy.bool_'> <type 'numpy.bool_'>

so if before it would cast numpy.bool_ into generic bool (which is imho more
preferable), now basic bool type gets casted into numpy.bool_, so
that v is not v_ any more

is that intentional change?

--
=------------------------------------------------------------------=
Keep in touch www.onerussian.com
Yaroslav Halchenko www.ohloh.net/accounts/yarikoptic

[Andrew Collette]

Hi,

> I wonder if that is intentional change introduced in 2.0.0 since in
> 1.3.0 it is all good:  consider following snippet:
>

> and on 2.0.0:
>
> original types  <type 'bool'> <type 'numpy.bool_'>
> read back type  <type 'numpy.bool_'> <type 'numpy.bool_'>
>
> so if before it would cast numpy.bool_ into generic bool (which is imho more
> preferable), now basic bool type  gets casted into numpy.bool_, so
> that v is not v_ any more

I don't think this was specifically intentional, but in general if a
NumPy type and native Python type both exist for a scalar quantity,
h5py should return the NumPy type. In this case numpy.bool_ is the
type I would have expected, if for no other reason than to make the
types obtained by slicing scalar and multidimensional datasets the
same. I think casting to native Python equivalents for bools was an
unintended side effect of some code in 1.3.1 that tried (poorly) to
support new types like variable length strings.

Andrew

[Yaroslav Halchenko]

Thank you for the clarification!

On Fri, 21 Oct 2011, Andrew Collette wrote:
> I don't think this was specifically intentional, but in general if a
> NumPy type and native Python type both exist for a scalar quantity,
> h5py should return the NumPy type. In this case numpy.bool_ is the
> type I would have expected, if for no other reason than to make the
> types obtained by slicing scalar and multidimensional datasets the
> same. I think casting to native Python equivalents for bools was an
> unintended side effect of some code in 1.3.1 that tried (poorly) to
> support new types like variable length strings.

> Andrew